Looping over a subset of times in the FieldSet

[gmacgilchrist]

I wish to evolve particles in a velocity field that loops multiple times over a specific year (for example, a year with certain anomalous conditions). I thought I would be able to do this by setting time_periodic to timedelta(365) and setting the initialization time to be the start of the year that I wish to loop. However, this fails with:

ValueError: Time series provided is longer than the time_periodic parameter

The implication is that I can only loop when I wish to do so across the time boundaries of the FieldSet.

Is there a specific reason for this constraint? Would it be possible to remove it, and thus have the potential to loop over a subset of the FieldSet?

I am running with parcels version 2.3.0.

[erikvansebille]

Thanks for bringing this up, @gmacgilchrist. I agree that this is not very clear in the documentation.

The design philosophy of the time_periodic functionality is that the looping will be over the full time-range of the FieldSet. If a user wants to run over a subset in time, they will have to make sure that the FieldSet is shortened (e.g., by providing fewer input netcdf files).

The reason why Parcels still supports a timedelta object in time_periodic is that Parcels doesn't know what the time interval is between the last snapshot (t_n) and the first snapshot (t₀). Parcels could assume that it's the same as between e.g. t₀ and t₁, but we want to keep the flexibility to support FieldSets with variable time intervals.

So, this is not a bug but a design choice. I will transfer this Issue to a Discussion

[gmacgilchrist]

Thanks for the response @erikvansebille, this makes sense, and I agree with retaining flexibility.

I guess what I am wondering is whether it is possible to implement a feature of being able to loop within the timeframe of one of the netcdfs provided to the FieldSet. To offer my example, I have 6 years of data (2013 to 2018) for the BSOSE data assimilation. The data are arranged into single files containing all 6 years, with one file for each variable. I would like to loop over just 2014 (a year with an anomalously strong Ross Gyre) and compare with trajectories looped over the full time-evolving field. To do this in the current set-up, I have to extract the single year (2014) from the main netcdf file, save it in a new netcdf file, and use this as my FieldSet. Obviously this is not a problem in the grand scheme of things, but one could feasibly imagine someone wanting to do something similar for, say, every year of a 60+ year hindcast, where the data are saved in 5-year intervals, in which case they would have to resample all of their data to annual output.

I'm wondering if there is a reason why something like that can't be performed in parcels?